Blogs > The Law Blogger

The Law Blogger is a law-related blog that informs and discusses current matters of legal interest to readers of The Oakland Press and to consumers of legal services in the community. We hope readers will  find it entertaining but also informative. The Law Blogger does not, however, impart legal advice, as only attorneys are licensed to provide legal counsel.
For more information email: tflynn@clarkstonlegal.com

Thursday, May 11, 2023

Machine Learning Fail Leads to a Pair of Landmark Cases in Michigan

Insurance litigation and government benefits law are some of the driest topics on the planet. But the two
Bauserman cases decided by the Michigan Supreme Court raise the timely issue of the role of artificial intelligence in making determinations of human intent; weighing evidence; and deciding between two versions of the same events. 

Machines Determining Benefits

The problem arose when the Unemployment Insurance Agency [UIA], the government agency that processes Michigan's unemployment claims, eliminated human decisionmaking in the determination of unemployment benefits. UIA adopted a machine learning tool known as the Michigan Integrated Data Automated System (MiDAS). This program played a significant role in the Bauserman cases.

MiDAS is an automated system used by the UIA to detect fraud in unemployment claims. In 2013, the UIA implemented a new algorithm in MiDAS that automatically flagged certain claims as potentially fraudulent, triggering a review by a computer program rather than a human investigator. Their experimentation with AI was very short lived; the state stopped using MiDAS as its sole method of detecting unemployment fraud in 2015.

In just over two years, however, the damage had been done. In addition to detecting when unemployment claimants committed fraud in their applications, MiDAS also imposed quadruple penalties upon applicants it determined were intentionally fraudulent. Perhaps the straw breaking the camel's back was that MiDAS interfaced with the state and federal treasuries to claw back wrongfully paid benefits from a so-called fraudulent applicant's tax refund.

A report issued by the state's auditor general concluded that MiDAS incorrectly decided 92% of its cases. The result was that tens of thousands of applicants had their unemployment claims wrongfully denied by a machine learning program. The machine cost the state millions and led to the class action suit in Bauserman.

The Bauserman Cases

Grant Bauserman was one of the many claimants who were affected by the new MiDAS algorithm. After Bauserman filed his claim, MiDAS flagged it as potentially fraudulent based on the new algorithm; his benefits were denied.

Bauserman was fired from his job as a bartender after he got into an argument with a customer and used profanity. He applied for unemployment benefits, but his claim was denied by the state's Unemployment Insurance Agency because it determined that Bauserman was fired for misconduct.

The bartender appealed the decision, arguing that the state's definition of misconduct was too broad and could include minor infractions. The case eventually made its way to the Michigan Supreme Court, which ruled in favor of Mr. Bauserman.

In the first Bauserman case, the Michigan Supreme Court addressed when an applicant's claim must be filed in the Court of Claims. The second Bauserman case involved whether the MiDAS computerized denial of unemployment benefits impacted a constitutional right of the class members.

The court held that the state's definition of misconduct was too broad and that claimants could not be disqualified from receiving unemployment benefits unless they engaged in willful and wanton misconduct. The court also noted that the burden of proof was on the employer to show that the employee's actions rose to the level of willful and wanton misconduct. Here is a link to the Bauserman II case dealing with an state benefit applicant's constitutional rights.

Ultimately, this class action case settled last September for $20 million. Eligible claimants who had their property seized by the state can apply for compensation from the settlement fund. The settlement still needs to be approved by the trial court; the next hearing on the matter is scheduled for July.

The Nomenclature

These cases are interesting because they involve "artificial intelligence" and "machine learning". Artificial intelligence (AI) and machine learning (ML) are related but distinct concepts in the field of computer science.

AI is a broad field that encompasses many different technologies and approaches, while machine learning is a specific approach to building AI systems that relies on data-driven algorithms and statistical techniques.

AI refers to the development of computer systems that can perform tasks that would normally require human intelligence, such as recognizing speech, making decisions, and learning from experience. AI involves the development of algorithms and computational models that can reason, perceive, and understand the world in a way that is similar to human cognition. This is what MiDAS attempted to pull off.

Machine learning, on the other hand, is a subset of AI that focuses on developing algorithms that can automatically learn and improve from experience without being explicitly programmed. ML algorithms use statistical techniques to analyze data, identify patterns, and make predictions or decisions based on those patterns.

The goal of machine learning is to enable computers to learn from data and improve their performance over time, without the need for explicit instructions or intervention from a human programmer. MiDAS may have been learning from its high rate of mistakes, but financially strapped unemployed workers were being financially penalized on the machine learning curve.

The Bauserman cases are also of interest because they involve machines wrongfully affecting people's lives. The cases highlight concerns about our state government relying too-heavily on machine learning tools. The UIA's fateful decision to rely solely on a computer program to determine benefit eligibility was disasterous, harmful and expensive. 

Next Steps

This will surely not be the last attempt by governments to replace the human decision making process with computers. In fact, these cases establish the fact that governments are already identifying and implementing machine learning tools.

Until a piece of code can "get it right" at least 92% of the time, humans will continue to play a key role in the determination of state benefits. Machine interference with the process is unconstitutional says our Supreme Court.

Clarkston Legal

Post #632

Labels: , , , , ,